Face Recognition

This project uses supervised learning to identify who is who in images of faces.

First, faces in images are identified and cropped. A histogram equalisation is performed, and the resulting image is scaled to 512x512 pixels. Then, it is passed to PCA algorithm to reduce dimensionality (each pixel is a feature).

Finally, different classification algorithms are used. Models are trained using 3-fold cross-validation. The goal is to provide and image to the model, and the model must say who is the person in that image.

Environment used:

• Python3.7 + Anaconda
• See requirements.txt

Structure:

• face_detector/ - Source code.
• faces/ - .jpg files inside A- Z folders. Dataset used: http://www.vision.caltech.edu/Image_Datasets/faces/faces.tar
• cascades/ - .xml cascades files for feature recognition. See more in http://alereimondo.no-ip.org/OpenCV/34
• main.py - Just type python main to run experiments.

Images folder structure inside faces folder follows the folder structure according to file face_detector/utils.py.